llama_model_loader: loaded meta data with 54 key-value pairs and 658 tensors from gemma-4-26B-A4B-it-WIP/gemma-4-26B-A4B-it-Q7_K.gguf (version GGUF V3 (latest))
llama_model_loader: - type  f32:  392 tensors
llama_model_loader: - type  f16:    2 tensors
llama_model_loader: - type q8_0:  236 tensors
llama_model_loader: - type q6_K:   28 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = Q8_0
print_info: file size   = 22.03 GiB (7.50 BPW) 

====== Perplexity statistics ======
Mean PPL(Q)                   : 17147.252339 ± 340.529496
Mean PPL(base)                : 6617.791728 ± 107.205431
Cor(ln(PPL(Q)), ln(PPL(base))):  93.49%
Mean ln(PPL(Q)/PPL(base))     :   0.952076 ±   0.007435
Mean PPL(Q)/PPL(base)         :   2.591084 ±   0.019265
Mean PPL(Q)-PPL(base)         : 10529.460611 ± 243.296833

====== KL divergence statistics ======
Mean    KLD:   0.636821 ±   0.004695
Maximum KLD:  38.045353
99.9%   KLD:  18.258318
99.0%   KLD:   9.417603
95.0%   KLD:   3.477128
90.0%   KLD:   1.624385
Median  KLD:   0.050566
10.0%   KLD:   0.000166
 5.0%   KLD:   0.000018
 1.0%   KLD:   0.000000
 0.1%   KLD:  -0.000001
Minimum KLD:  -0.000211

====== Token probability statistics ======
Mean    Δp: -0.136 ± 0.027 %
Maximum Δp: 100.000%
99.9%   Δp: 95.973%
99.0%   Δp: 31.761%
95.0%   Δp:  4.454%
90.0%   Δp:  0.622%
75.0%   Δp:  0.001%
Median  Δp: -0.000%
25.0%   Δp: -0.003%
10.0%   Δp: -0.878%
 5.0%   Δp: -5.361%
 1.0%   Δp: -34.506%
 0.1%   Δp: -98.011%
Minimum Δp: -100.000%
RMS Δp    : 10.174 ± 0.094 %
Same top p: 77.572 ± 0.109 %